What changed
Since late August, cybercriminals have been using AI-powered penetration testing tools to launch attacks on companies, government agencies, and banks worldwide. These tools, which include frameworks like Artex, allow hackers to automate and accelerate cyberattacks that would otherwise take days to complete. The tools do not introduce new attack methods but enable faster execution of existing techniques.
What this means in practice
Artex, an open-source framework written in Go, has been linked to multiple real-world attacks. It acts as a guide for connecting external AI models to existing penetration testing tools, allowing users to issue simple natural-language prompts to initiate attacks.
For example, hackers have used Artex to attempt to gain access to a victim's admin panel, database, or server. Once instructions are given, multiple AI agents carry out the tasks, breaking down plans into actions and presenting findings.
However, safety guardrails in AI models can sometimes block hacking attempts, though increasingly, free Chinese and other models are bypassing these protections.
Limitations and availability
Artex is not an AI product itself but a framework that connects to external AI models via APIs. It was first published on GitHub in July by a pseudonymous developer using the handle Autumn-27.
Despite the developer removing Artex from GitHub, multiple versions continue to be used and developed. Security researchers have identified over 45 AI attack frameworks, with Artex being one of the most prominent.
These tools are currently easy to detect due to the noise they generate in systems, such as sending thousands of requests in a short time.
Basic security practices like patching software, avoiding default credentials, and monitoring network activity can help defend against these attacks. The supplied announcement does not specify the exact number of cybercriminals using Artex or other frameworks.
